Large Language Models for the Simulation of Human Behavior

Recent advances in large language models (LLMs) have unlocked new possibilities for simulating human behavior across a variety of domains. In computer science, LLM-based user simulation is increasingly applied to the evaluation and training of search engines, recommender systems, and conversational assistants, enabling scalable, controllable, and reproducible interaction testing without direct human involvement. Parallel to these developments, researchers in the social and behavioral sciences have begun to employ LLMs as synthetic populations, agents, or role-playing participants to study decision-making, group dynamics, persuasion, and other cognitive and social phenomena. This survey reviews and synthesizes the emerging literature on LLM-driven simulation of human behavior, spanning both computer and social science perspectives. We highlight publicly available toolkits and frameworks that facilitate reproducible research. We further discuss validity concerns associated with replacing or augmenting human data with LLM-generated data. Finally, we identify open research challenges, outlining a roadmap for future research in this field.